calculate one-step-ahead (prediction) residuals from a aniMotum
ssm fit
osar(x, method = "fullGaussian", ...)a aniMotum ssm fit object with class ssm_df
method to calculate prediction residuals
(default is oneStepGaussianOffMode; see TMB::oneStepPredict for details)
other arguments to TMB::oneStepPredict
One-step-ahead residuals are useful for assessing goodness-of-fit
in latent variable models. This is a wrapper function for TMB::oneStepPredict
(beta version). osar tries the fullGaussian (fastest) method first and
falls back to the oneStepGaussianOffMode (slower) method for any failures.
Subsequent failures are dropped from the output and a warning message is given.
Note, OSA residuals can take a considerable time to calculate if there are
many individual fits and/or deployments are long. The method is automatically
parallelised across 2 x the number of individual fits, up to the number of
processor cores available.
Thygesen, U. H., C. M. Albertsen, C. W. Berg, K. Kristensen, and A. Neilsen. 2017. Validation of ecological state space models using the Laplace approximation. Environmental and Ecological Statistics 24:317–339.
# generate a ssm fit object (call is for speed only)
xs <- fit_ssm(ellie, spdf=FALSE, model = "rw", time.step=24, control = ssm_control(verbose = 0))
#>
res <- osar(xs)